Research map / Video / Video segmentation and tracking
D3: Training-Free AI-Generated Video Detection Using Second-Order Features
Chende Zheng, Ruiqi Suo, Chenhao Lin, Zhengyu Zhao, Le Yang, Shuai Liu, Minghui Yang, Cong Wang, Chao Shen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use second-order instead of first-order feature std (ablation) | D3 · GenVideo | mAP: 95.69 -> 98.46 | Improved |
| use second-order instead of first-order feature std (ablation) | D3 · EvalCrafter | mAP: 86.40 -> 98.87 | Improved |
| use second-order instead of first-order feature std (ablation) | D3 · VideoPhy | mAP: 86.06 -> 99.16 | Improved |
| use second-order instead of first-order feature std (ablation) | D3 · VidProM | mAP: 80.61 -> 88.46 | Improved |
| use L2 distance instead of cosine similarity as first-order feature (ablation) | D3 XCLIP-B/16 · GenVideo | AP: 91.30 -> 97.72 | Improved |
| use L2 distance instead of cosine similarity as first-order feature (ablation) | D3 XCLIP-B/16 · VidProM | AP: 79.87 -> 87.08 | Improved |
| use XCLIP-B/16 instead of MobileNet-v3 encoder (ablation) | D3 · GenVideo | mAP: 95.47 -> 98.46 | Improved |
| use MobileNet-v3 instead of XCLIP-B/16 encoder for speed (ablation) | D3 · GenVideo | inference time (s): 56 -> 40 | Improved |
| use XCLIP-B/32 instead of XCLIP-B/16 (patch size) (ablation) | D3 · GenVideo | mAP: 98.46 -> 98.15 | No clear effect |
| use XCLIP-B/16 instead of MobileNet-v3 under Gaussian blur sigma=4 (ablation) | D3 · GenVideo | mAP: 84.38 -> 92.69 | Improved |
| use XCLIP-B/16 instead of CLIP-B/16 under JPEG q=60 (ablation) | D3 · GenVideo | mAP: 78.81 -> 94.50 | Improved |
| train-free second-order features vs trained Demamba on VideoPhy | D3 · VideoPhy | mAP: 51.47 -> 99.16 | Improved |
Closest papers by meaning
- Preserving Forgery Artifacts: AI-Generated Video Detection at Native Scale (ICLR 2026 · similarity 0.88)
- AI-Generated Video Detection via Perceptual Straightening (NeurIPS 2025 · similarity 0.87)
- Physics-Driven Spatiotemporal Modeling for AI-Generated Video Detection (NeurIPS 2025 · similarity 0.87)
- Seeing What Matters: Generalizable AI-generated Video Detection with Forensic-Oriented Augmentation (NeurIPS 2025 · similarity 0.86)
- Rethinking the Readout: Unlocking Video Backbones for AI-Generated Video Detection (NeurIPS 2026 · similarity 0.86)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
